In the documentation (https://www.rosettacommons.org/docs/latest/scripting_documentation/RosettaScripts/Filters/Filters-RosettaScripts), there's the following text states that the filter will change its behavior based on the confidence value. I'd like to ask what it means to be "probed" here, suppose I give a confidence of 0.99, does that mean that the filter will return True in 1% of the cases irrespective of what inputs it had, or that it will return True in 1% of the cases where it would otherwise return False?

If confidence is 1.0, then the filter is evaluated as in predicate logic (T/F). If the value is less than 0.999, then the filter is evaluated as fuzzy, so that it will return True in (1.0 - confidence) fraction of times it is probed. This should be useful for cases in which experimental data are ambiguous or uncertain.

"Probed" in the paragraph means "run" (or asked for a pass/fail decision).

The confidence is such that if the Filter returns (or would return) "pass", it *always* counts as a pass.

The confidence only comes into play if the Filter would return "fail". In those cases, the "fail" verdict is only obeyed the fraction of time given by the confidence setting. In the other (1.0 - confidence) cases, it's treated like the Filter returned a "pass" instead.

To avoid the runtime expense for complicated filters, the exact implementation actually inverts the order, so (1.0 - confidence) fraction of the time it always returns "pass", and (confidence) fraction of time it actually calculates the filter value and acts on it. The actual effect should be the same as above, though.